Binocular Stereoscopic Vision Positioning Technology Based On GPU

At present,the visual system has been widely used in the fields of intelligent detection system and intelligent equipment,and has become an important standard to measure the performance of intelligent equipment.As one of the key technologies of the intelligent equipment vision system,the target positioning technology has always been a hot topic in the research.Traditional target location technology relies on monocular vision and obtain less information.Binocular stereo vision technology can get distance information from a pair of corresponding images,but because of its complex computation,it is not suitable for real-time operation.With the advent of CUDA,the parallel data processing capability of GPU has been gradually paid attention?Therefore,in this subject,we combine the binocular vision with GPU for target distance measurement,and we called it a binocular vision stereo positioning technology based on GPU?First of all,a model of binocular vision stereo measurement system is built.We need to analyze the hardware requirements and precision requirements of the measurement system at first,and then design the implementation process and hardware selection of the measurement system,according to the actual environment of the system application.At last,the binocular vision stereo measurement system is designed and implemented in the VS2013,based on the open source computer vision library Open CV.Secondly,in view of the actual application scene,the binocular vision target detection is realized by using the color space conversion.For the complex background of the actual project,we use the color characteristics of the logo in the image to identify the target,and through the transformation of color space,we can get the contour of the logo in the image,and reduce the external interference.Then the ellipse center point coordinates are worked out using the minimum circumscribed,and the three-dimensional distance of the logos is calculated by using the central pixel coordinate and the optical axis convergent measuring principle in the binocular stereo measurement system.Then,aiming at solving the complex problem of binocular vision computing,the parallel method of location algorithm is studied.The CUDA programming model and the memory model are deeply studied,and the image algorithm and the 3D coordinate calculation are accelerated on the CPU+GPU heterogeneous platform,the parallel tasks of the algorithm are divided into detail,and the implementation of CUDA platform programming is implemented,the experimental results show that the system achieves 6 times of acceleration through CUDA and meets the real-time requirement of the system.Finally,a binocular vision stereo positioning platform is built to verify the implementation of the system.The performance of the algorithm is experimentally analyzed under different conditions,and the feasibility of the algorithm is verified.The experimental results show that the proposed system implementation method can accurately detect the target and achieve the three-dimensional distance calculation of targets,and the measurement accuracy is within the range of + 10 mm,and the computation speed is 141.6ms/pic.